摘要
针对传统的增式支持向量机算法在计算时间和分类效率上的不足,提出了一种新型的增式SVM训练算法。该算法不是简单地保留上一步训练的支持向量,而是通过增加KKT(Karush-Kuhn-Tucke)限制条件并对决策函数的输出设定一个阈值,使得保留下来的样本都是最有效的样本,从而可减少训练样本的数目。在仿真实验中,选择了一组UCI数据,并选用RBF核函数作为核函数。实验结果表明:与传统增式算法相比,新算法在保证传统SVM性能的同时,在迭代速度和分类放率上分别提高了14%和4.39%。
For the insu time and the serves the su efficiency ffic of iencies of original incremerital algorithm of support classification, this paper proposed a new incremental vector machine in operation algorithm: it is not only repport vectors of the former training step simply, but also adds KKT(Karu condition as the restrict condition, and sets a threshold to the decision function's output. served samples the most effective. The kernel function as the kernel function. result indicated that the new in but also can improve 14% and crementa sh-Kuhn-Tucke) It makes the reexperiment data are chosen from UCI database, and choose RBF According to the original incremental algorithm, The experiment 1 algorithm not only can conserve the capacity of the original SVM, 4.35 % on operation speed and classification efficieng.
出处
《青岛大学学报(工程技术版)》
CAS
2007年第3期82-85,共4页
Journal of Qingdao University(Engineering & Technology Edition)
关键词
支持向量机
增式训练算法
KKT条件
support vector machine
incremental training algorithm
KKT condition